Compare it with a micro-cached Varnish or nginx .. caching is outside of the "speed" parameters. Of course, you could take a look memcached if you want :)
I can't find the reference I remember but it fits roughly with the link posted (the scale seems to match the max times rather the average times). The differences were tiny but the order was PHP, Python, Ruby if I remember correctly (PHP was in front, PR may have been switched).
In the case of Mako at least, caching doesn't mean caching in the way you mean it. It's more akin to opcode caching in PHP, saving the actual python code generated from the template, not the result. Just saves from reparsing the template every time it's rendered, but the actual template logic still runs.